In traditional search, you can be ninth. Ninth is a bad place to be, but it is a place. Someone who scrolls far enough will find you, and a rank tracker will tell you that you moved up from twelfth.
AI search does not work that way. Ask ChatGPT or Perplexity for the best birding blogs and you get an answer with three or four names in it. You are one of those names or you are not. There is no page two.
That changes what a web strategist has to measure. I learned it on a side project.
The audit
In June I launched Trailside Birding, a blog that picks the best birding trails in a region. Each guide pairs trail details from AllTrails with species data from eBird for the same spots.
Three days after launch I ran an AI visibility audit. The method is simple. Write down the questions a real person would ask. Run them through ChatGPT, Claude, Gemini, and Perplexity. Record who gets named. I tested prompts like these:
- "Best birding trail guide blog"
- "Best NYC birding guide"
- "Kirtland's Warbler tours Michigan"
- "Pacific Northwest birding trails"
- "Birding guide that combines AllTrails and eBird"
Then I scored the site on presence, position, and how completely the answers described it.
It scored 8 out of 100. That was generous. The site was three days old and appeared in zero answers.
Who did show up
The general questions belonged to the organizations you would expect. Cornell Lab's All About Birds, Audubon, 10,000 Birds, and state wildlife agencies filled almost every answer. They earned that with decades of content and thousands of sites linking to them.
A three-day-old blog is not going to take "best birding blog" from the Cornell Lab of Ornithology. It should not try.
The question nobody was answering
One prompt came back different. When I asked for a guide that combines trail data with bird data, no single site owned the answer. The AI stitched one together from eBird hotspot pages, AllTrails listings, and general travel advice.
That was the finding worth the whole exercise. A birder planning a trip has two questions. Where are the birds? Can I actually walk there? The big birding sites answer the first. Trail apps answer the second. Nobody was answering both in one place.
So the strategy stopped being "rank for birding." It became "be the obvious answer to the combined question."
What that changes in practice
Say what you are in one plain sentence. AI tools describe you with language they find about you. A vague About page gets you a vague description. Ours now leads with the exact question the site answers: where do the best trails and the best birding overlap?
Frame every page around that question. Titles, descriptions, and opening paragraphs should make the combination obvious. Covering both halves somewhere in the body is not enough.
Get mentioned elsewhere. What other sites say about you counts for more than what you say about yourself. That means being useful in the places birders already gather.
Do the SEO basics first. AI tools that search the web can only cite pages that are indexed. Sitemaps, structured data, and clean metadata still come first. GEO is built on top of SEO, not instead of it.
The prompt list is the new keyword list
For an established organization the questions change. You are probably already named. What matters is whether the description is accurate, where you land in the list, and whether the answer repeats something you stopped saying three years ago.
The method works at any size. Write down the twenty questions your audience actually asks. Run them every month across the major tools. Track whether you are named, where you appear, and whether what is said is true. That list belongs next to the keyword list in any web strategy, and it costs nothing to start.
Four months later
In October I reran the prompts in Gemini, ChatGPT, Perplexity, and Claude. Every run was in a private browser window. Gemini and ChatGPT were logged out. Perplexity was signed in to an old account. Claude requires an account, so that run was signed in too.
- "Birding guide that combines AllTrails and eBird." First in Gemini, ChatGPT, and Perplexity.
- "Best NYC birding guide." First in Gemini, ChatGPT, and Perplexity.
- "Best birding trail guide blog." First in Gemini and Perplexity. Fourth in ChatGPT.
- "Pacific Northwest birding trails." First in Perplexity, which named all three trails in the guide. Not mentioned in Gemini or ChatGPT.
Claude searched the web on every prompt and did not mention Trailside on any of the four.
On the combined question, three of the four tools answered with the site's name before anything else. Gemini opened with "The exact resource that matches this description is Trailside Birding." ChatGPT opened with "The closest match is Trailside Birding." That is the question the whole strategy was built around, and in July nobody owned it.
The trail guide prompt is the one I did not expect. In July it belonged to the big birding sites. Now Trailside is first in Gemini and Perplexity and fourth in ChatGPT.
The Pacific Northwest prompt is the honest counterweight. Ask about a region instead of a format and three of the four tools skip Trailside entirely. A broad regional question still goes to the organizations with years of regional content. The site wins where it answers a question nobody else answers, and not much further than that yet.
Claude is the other counterweight. Four for four in Perplexity and zero for four in Claude, on the same day, for the same site. Each tool searches and weighs sources its own way. Doing well in one tells you very little about the others, which is the strongest argument I know for testing all of them instead of the one you happen to use.
Being named is not the same as being described correctly
Every tool that named the site described it the way the site describes itself: trail data and bird data in one guide. Gemini repeated the NYC trail distances and eBird species counts correctly. Perplexity named the right three Pacific Northwest trails.
ChatGPT got one thing slightly wrong. It called the posts "trip reports." They are regional guides, not accounts of trips I took. That is a small miss, but it is exactly the kind of drift you only catch by reading the answers instead of counting mentions. At a larger organization, the same drift might be an outdated program name or a retired statistic.
This is four tools, on one day. AI answers vary from run to run. The site now has 16 regional guides, and I will keep running the same prompts.
The July audit projected a score of 45 to 60 out of 100 by January. I still would not put much weight on a number like that. "Named first for the question you set out to answer" is a better measure, and it is the one that shows up when someone asks.